[Purposes] In clinical practice, professionals need to analyze and diagnose the electrocardiograph (ECG) beat by beat, which is time-consuming and energy-consuming. To address this issue, an automatic ECG identification method based on a pre-trained Inception network was proposed. [Methods] Firstly, the Mexican wavelet transform was used to convert the ECG from the time domain to the time-frequency domain and extract the time domain and frequency domain information of heartbeat signals. Secondly, the Inception network was utilized to automatically diagnose and identify time-frequency graphs of heartbeats, and the stochastic gradient descent momentum (SGDM) algorithm was adopted for model optimization during the training. [Findings] In order to verify the effectiveness of the proposed method, five types of heartbeat data from the public arrhythmia database were selected, and experimental results show that the proposed algorithm performs well in indicators such as positive predictive value, recall rate, and accuracy, and it has higher precision and faster convergence compared with the pre-trained residual networks and visual geometry group networks under the same experimental conditions. [Conclusions] The Mexican wavelet basis function can better characterize the shape of a single heartbeat, and the end-to-end Inception model can concatenate the heartbeat signal feature matrices with different widths according to the depth and extract richer features.